HotelME

Use this when comparing 2–3 returned public hotels in input order; no Metro login. Dates are optional for aggregate review comparisons; do not block them on dates. Compare the returned hotels in one call rather than parallel per-hotel detail calls. Preserve unknowns. Theme counts overlap, not unique guests. Display date-prefilled URLs verbatim; observed price is not a guaranteed offer at the URL. Dates and party must not be altered to bypass unsupported conditions. No reservations or live quotes. State the actual displayed hotel count. Include each displayed price's OTA source, source_observed_at and differing price bases, not just the cheapest amount. Repeat source/date/time/timezone in each follow-up comparison or changed-date answer, even if already stated earlier. Use table columns or a note explicitly bound to the relevant hotel/date; a shared note is valid only when those provenance values match.

Use this when reading aggregate review evidence for one returned public hotel. Dates are optional for aggregate analysis; never expose raw guest text. Reuse an existing result if it already contains the requested evidence, prices or links. Pass exact known dates/party for booking searches; never return a hotel front page instead. dates_required applies only to booking links, not aggregate analysis; ask dates only if links or prices are requested. Missing links never permit guessing. Show the source and source_observed_at of any displayed observed price. No Metro login. Cannot provide unverified amenities, walk times, photos or private reports. Cannot bypass unsupported-party limits from another tool.

Use this when comparing cached daily prices for 1–3 hotels across 1–14 nights. Dates are required for this calendar tool; not for review-only discovery. No login. Check-out is exclusive, within today JST to six calendar months. The entire stay must fit returned supported_window: start <= check_in < check_out <= end. Do not predict future data readiness from check-in alone. Only 2 adults, 1 room, no children. Request a flexible one-night range once, not per day. Compare hotel×date combinations, retaining repeated hotels unless unique hotels were asked. Unknown/stale/uncollected is not zero or sold out. Source sell-out is not hotel-wide fullness. Observed prices are not live quotes, same-plan stay totals or tax guarantees. Use source_availability_labels per OTA/date, combining observations with source_availability_gaps. For each source_availability_sources entry missing from both, use source_availability_default (not_collected). Never infer sold-out from absence. Match source aliases: じゃらん=Jalan.net, 楽天トラベル=Rakuten.co.jp. Keep unknown, no usable observation and expired information distinct. Show an expired observation's time only as historical, never its price. Missing source records do not establish a failed fetch or no previous collection. Show every displayed source availability status with source_observed_at and timezone, or 取得日時不明 when absent. Include each displayed price's OTA source and source_observed_at. If the source time is unknown, say so; checked_at is a response check time, not an OTA observation time. Show supplied booking URLs verbatim using normal clickable Markdown links, not button-only controls; never invent missing links.

Use this when mapping a review preference to published theme IDs; no Metro login. For a hotel search with an ambiguous destination (e.g. 港区/中央区/北区), first ask which prefecture/city unless the user's explicit context resolves it; never assume Tokyo. A request for 港区 alone needs clarification (Tokyo, Osaka or Nagoya); a theme, language, device location or a popular hotel cannot resolve that ambiguity. This takes priority over choosing breakfast subtypes. Standalone theme browsing needs no destination. Discover once per keyword and reuse IDs; dates are not needed. General breakfast is not buffet; onsen does not imply a private bath. Review mentions are aggregate analysis, not verified hotel facilities or individual guest text. Keyword matches are alternative topics, not a list of jointly required filters. For good breakfast, use a returned general breakfast theme when available. If only narrower subtypes are returned, ask one brief clarification before searching. Do not select buffet or another subtype without the user's consent.

Use this when finding hotels by name, destination or aggregate review themes. Resolve ambiguous destinations (e.g. 港区/中央区/北区) from explicit user context or ask which prefecture/city before looking up themes or asking about subtypes. Never assume Tokyo or another city; do not invent a prefecture. Example: 港区で眺めのいいホテル needs a city clarification, not a Tokyo search. Offer 東京都港区, 大阪市港区 and 名古屋市港区, not just Tokyo and Osaka. 大阪市港区で眺めのいいホテル is explicit: preserve Osaka, never substitute Tokyo. For bare 品川 ask 品川駅周辺 or 品川区. Station areas and city/ward boundaries differ. An administrative-area match never verifies distance from a station. If a station or walking-radius constraint cannot be verified, ask before broadening the area; do not silently search 博多区 for 博多駅周辺 or label such results station-near. Dates are optional for hotel discovery and aggregate review analysis. Run these searches without dates; do not require a stay date just to find breakfast themes. Ask dates only when prices, budget filtering or date-prefilled booking links are requested. Metro registration/login is not required. Return up to 5 hotels. Pass all known dates, party, themes and budget together. Every selected theme must match: include only explicitly required themes. For good breakfast, use a returned general breakfast theme when available. If only narrower subtypes are returned, ask one brief clarification before searching; do not select buffet without consent. Related themes can be comparison dimensions, not mandatory search filters. Put the requested city/ward in location and hotel/chain name in hotel_name; do not merge them. A prefecture alone does not narrow to its similarly named city. Keep branch/building/sub-brand words in hotel_name. Location is not a meal preference or walking-distance guarantee. Japanese and English city forms are accepted. For explicit exclusions, pass prior returned IDs in exclude_hotel_ids or the user's named hotel/brand keywords in exclude_hotel_names. Exclusions run before ranking and budget selection, so eligible replacements can fill the requested limit. Preserve exclusions on follow-up searches along with all other conditions. Name exclusions are not a corporate ownership or entire hotel-group guarantee: do not infer unnamed affiliated brands. Clarify a requested corporate group if its brands are unknown. Names allow width/kana/separator variants and curated aliases, not arbitrary typos. If search_diagnostic is returned, explain its reason. name_suggestions are only possible identities, NOT matching recommendations; ask which hotel was intended. name_search_limited is an inconclusive bounded search, not proof that a hotel does not exist or lacks review evidence. Missing names can reflect catalog coverage or a different registered name. Never replace an unavailable named branch with a similarly named sister hotel or claim it is that same property. They have no verified theme/price match. After confirmation, retry recommend_hotels with the confirmed name and ALL original place, branch, theme, date, party and budget conditions. Do not bypass these filters by fetching suggestion IDs through details. A limited/unavailable name lookup is not proof that the hotel is absent. Zero results do not prove missing data, poor reviews or sell-out; clarify the name or city without dropping other conditions. Show returned registered names. Ask for destination when a vague trip is unresolved; do not substitute national leaders. No photo URLs are returned. Prefer text and clickable links; omit image carousels and unavailable-image placeholders. External citations must match the exact branch and location, not merely the same brand; omit unverified factual claims. Report the actual number of hotels displayed, not the requested limit. Do not count name_suggestions or excluded locations as matches. If fewer match, state that count. Review-count order is not overall quality. Nightly and whole-stay budget caps are JPY per room, not per person; never divide total budget by nights. Both caps apply if set. Budget filtering must occur before top-k; unknown/stale prices cannot satisfy a cap. A sum of observed nightly minima is not a same-plan quote or live availability. Prices and date-prefilled booking searches support 2 adults, 1 room, no children only. Never change party/dates, silently relax conditions or invent prices/links. Show supplied booking URLs verbatim, including dates. Use normal clickable Markdown links, not button-only controls. No separate lookup for the same links or prices already returned here. An unsupported condition requires asking the user before continuing, not a retry through another tool. No booking/payment actions. Include each displayed price's OTA source and source_observed_at; an unknown source time must be labeled unknown, not replaced with response checked_at. Show every displayed source availability status with source_observed_at and timezone, or 取得日時不明 when absent. Stale timestamps are historical only; never show expired prices. For not_collected, do not invent a failed-fetch reason or claim it was never collected. Search one condition set at a time and reuse returned results; do not fan out repeated searches.